hanes machin november 2012 0 Hate Crime in the Wake of Terror Attacks: Evidence From 7/7 and 9/11 Emma Hanes* and Stephen Machin** November 2012 * Department of Economics, University College London ** Department of Economics, University College London and Centre for Economic Performance, London School of Economics Abstract This paper asks a straightforward question - what happened to racially motivated hate crimes in the wake of the 7/7 terror attack that hit London in July 2005 and the 9/11 terror attack that hit the US in September 2001. There is anecdotal and statistical evidence of an increase in bias-motivated crimes since the 9/11 terrorist attacks in the US, but no quantitative research that has accurately pinned down the magnitudes of any hate crime increase that ensued. The study provides a unique estimate of the magnitude and duration of the effects of 7/7 and 9/11 on hate crime using data from four police force areas in England with sizable Asian/Arab populations. We find significant increases in hate crimes against Asians and Arabs that occurred almost immediately in the wake of both terror attacks and which lasted for a prolonged period. Moreover, hate crimes against Asians and Arabs do not return back to their pre-attack levels, showing a permanent increase in the wake of the attacks. We argue that this demonstrates a strong link between terror attacks and increases in hate crime and hypothesise that attitudinal changes from media coverage act as an underlying driver. JEL Classifications: K42. Keywords: Hate crimes; Terror attacks. Acknowledgements We acknowledge helpful comments and suggestions from Mirko Draca and from participants in a CEP seminar, the NBER Summer Workshop Economics of Crime session and at the 4th Annual IZA Meeting on the Economics of Risky Behaviors in Istanbul. 1 1. Introduction A small, but growing, literature has studied empirical issues surrounding the economic and social effects of terrorism. Attempts have been made to quantify the effects of terrorism on a number of outcomes, including GDP (Abadie and Gardeazabal, 2003; Bloom, 2009), financial markets (Chen and Siems, 2004), social attitudes or well-being (Bozzoli and Mueller, 2009; Frey, Luechinger and Stutzer, 2004), birth weight (Eskenai, et al, 2007; Lauderdale, 2006; Smits et al., 2006) and mental health (Metcalfe, Powdthavee and Dolan, 2011). Perhaps surprisingly, the evidence seems to suggest that the total effect on GDP and financial markets of a single terrorist incident is relatively short-lived, while the effects on well-being and health are large and persistent. In this paper, we explore a different question, what happened to racially motivated hate crimes in the wake of the 7/7 terror attack that hit London in July 2005 and the 9/11 terror attack that hit the US in September 2001. This is an interesting outcome to study if, for whatever reason, terror attacks alter individuals' perceptions of other groups in society. The paper empirically models the impact of terror attacks on hate crimes, in a setting with a credible research design where focus is placed on the impact on a particular sub-group of society. To do so, we investigate what happened to hate crime against Asians and Arabs in four regions of England after the 9/11 attack in the US in 2001 and the 7/7 attacks that hit London in 2005. Whilst there is anecdotal and descriptive evidence of an increase in hate crimes against Muslims since the 9/11 terrorist attacks (which we review below), we are not aware of much quantitative research that tries to accurately pin down the impact of terror attacks on the incidence of hate crimes.1 This is what we offer in this paper, where 1 There is a small amount of related work in sociology. Disha, Cavenish and King (2011) look at FBI data on hate crime in the US before and after 9/11. Deloughery, King and Asal (2012) study the direction of 2 we analyse rich monthly data before and after the terror attacks in four English police force areas with a significantly sized Asian/Arab (predominantly Muslim) population. We quantify the increased number of hate crimes against UK Muslims that occurred as a result of both the 9/11 attacks and the 7/7 bombings, using data that sub- divides hate crimes by victim ethnicity. Thus, we can study hate crimes against Asians and Arabs before and after the attacks, and generate estimates by using hate crimes against Blacks and Whites as control groups. One clear advantage of studying hate crimes as recorded by the English police is that they are explicitly defined and quantifiable. This therefore facilitates accurate study of time trends in a way which is not possible with the kind of opinion survey attitudinal data, self-reported well-being or newspaper coverage that have been more commonly studied in the terrorism literature. Moreover, hate crimes have greater implications – there is a direct cost to the victim, which may not be the case with attitudinal changes. To preview our main findings, we report sizable increases in hate crimes against Asians and Arabs - of the order of 20 to 30 percent - that occurred almost immediately in the wake of the two terror attacks. Moreover, the increase persisted and lasted for a prolonged period. In the case of the 7/7 attacks, where we have better data to estimate duration effects, increases remained on the margins of statistical significance for more than two years after the attacks and remained around 10 percent higher than the pre- attack levels, suggesting a permanent increase as a result of the terrorist attack. The structure of the rest of the paper is as follows. In Section 2, we consider some theoretical background motivation of our questions of interest and discuss relevant existing evidence. In Section 3, we describe the data we use and offer some initial causation between hate crime offences and terrorism, arguing strongly that hate crimes occur in response to terror attacks, but that there is no evidence of causation in the other direction where hate crimes would act as a precursor to terrorist activity. 3 descriptive analysis. Section 4 explains the modelling approach and presents statistical estimates of the impact of the 9/11 and 7/7 terror attacks on hate crime. Section 5 concludes. 2. Theoretical Background and Existing Evidence Hate Crimes in the Economics of Crime Becker's seminal (1968) paper was the first to consider crime in an economic framework of rational behaviour. According to his theory, agents maximise utility by comparing the benefits of crime with the costs, where costs are the time and effort required plus the expected cost of deterrence efforts (i.e. cost of a police fine or incarceration multiplied by the probability of detection and prosecution). Thus, Becker considers crime as a simple cost-benefit choice, and the model generates clear empirical predictions about incentive and deterrence effects on crime. In the original Becker model, harm or loss to the individual is considered an externality, essentially an unintentional side effect of the offender's actions. In the case of a hate crime, however, it has been suggested that loss to the victim is the intention of the crime (Gale, Heath, and Ressler, 2002; Craig, 2002). As well as causing harm to the victim, a hate crime is often intended to convey a message to the wider group to which the victim belongs (or is perceived to belong). Gale et al. (2002) thus extend the individual economics of crime model to a setting where a person’s utility function can contain a function of another person’s utility. They argue that one can understand racist acts and racial bigotry using this model. An individual sorts the world into groups of people, choosing to ‘like’ and ‘dislike’ these groups according to various characteristics. Thus, we can imagine that 4 the individual may choose to commit a hate crime in the case where the utility gained from seeing the ‘disliked’ person hurt outweighs the cost of effort plus expected loss from being caught and punished.2 An alternative, related, model of hate crime has been developed by Medoff (1999). His model proposes that an individual gains utility from two sources; hate crimes and all other goods. The crucial assumption is that, while other goods can be purchased on the market, hate crimes must be 'created' using personal time and resources, and are therefore a more time-intensive consumption activity. As a result, an event which causes the individual to value his or her time more highly (for example, an increase in wages) results in substitution away from hate crime activities and towards market goods. In both of these approaches, hate crimes can be viewed as a consumption good that generates utility, but at the same time incurs some kind of cost. In this setting, hate crimes could be driven by factors that alter preferences, for example if the propensity to commit hate crimes is affected by some kind of shock. One can ask what kinds of shocks may occur that could make an individual choose to dislike a hated group more or less at different times? At the micro level, this may be about personal experiences, education, culture and environmental changes. At the macro level, however, we might expect the biggest driver to be current affairs. So in the specific context of the hate crimes we study, namely those targeting Muslims, news events which some individuals may interpret as showing Muslims in an unfavourable light could be expected to increase the incidence of hate crime. We could plausibly consider the 9/11 and 7/7 terrorist attacks we study as featuring an extreme form of this media exposure. 2 This, of course, bears similarities to another area of Becker's (1957) work, namely that on tastes for discrimination. In this sense, as with the case of employers, workers or consumers having a taste for discrimination, one could think of individuals having a taste for hate crimes. 5 Hate Crimes in a Behavioural Approach So far, we discussed hate crimes within the economist's rational decision-making framework. When an individual decides to commit a hate crime, they do so because the expected utility from the action is positive. An alternative view is offered by contributions from behavioural economics. Particularly relevant are those areas which try to understand why agents make seemingly irrational decisions, even once factors such as limited information and limited decision-making time have been taken into account. It seems reasonable to think of hate crimes in this context in that, whilst the prospective gains from acquisitive crime are self-evident, the potential 'gains' from committing a pure act of violence against others are less clear (unless people have a taste for discrimination of this sort, though ultimately this is a theoretical proposition that is hard to test in practice). An alternative perspective might consider a hate crime to be an action of passion or emotion – where feelings of anger and rage dominate the individual's rational decision-making process. This is the assertion of Gordon and Arian (2001) who claim that “when one feels very threatened, the decision-making process is dominated by emotion rather than logic or rational considerations” (Gordon and Arion, 2001, page 197).3 Indeed, unlike other types of violent crime, hate crimes tend to be committed by groups of people rather than individuals (see Craig, 2002). This suggests there may be some element of group interaction, such as peer pressure or removal of social barriers, which causes individuals to commit hate crimes only when in groups. The concept of 'herding' is well known to economists, in particular in relation to financial markets. For example, economists explain the formation of stock market bubbles as being caused by 3 Gordon and Arion (2001) try to demonstrate this point using opinion poll data from Israel and America. The basic finding from the Israeli survey is that the more threatened by Palestinians the respondents feel, the less likely they are to support the establishment of a Palestinian state. 6 investors valuing assets according to how they believe others to value assets rather than based on private valuations. This kind of group behaviour can lead to seemingly irrational choices and can cause instability in financial markets (Baddeley, 2010). In the context of hate crime, we can imagine that group mentality has the power to overcome social taboos or persuade individuals to commit acts they would not otherwise have considered in order to impress the group. Escalation may occur when group members second guess the value that other members place on committing hate crimes. How do these notions connect to terror attacks? It is evident that a terrorist attack can trigger sharp changes in behaviour, which may not be rational responses (see Viscusi and Zeckhauser, 2003, or Sunstein, 2003). However, the supposedly irrational 'certainty premia' phenomenon is accounted for in a rational framework developed by Becker and Rubinstein (2009). They argue that, when considering shock mass-fear type events, the standard state-dependent utility model is not sufficient. In fact the model they develop assumes that a negative utility shock occurs only in a 'bad' state (like when the terrorist attack occurs), and not in good states. Thus, there are both rational and behavioural arguments that have been proposed to explain why hate crimes occur. In terms of empirical analysis, testing the distinction between the rational and behavioural arguments is not within the scope of this study (and it is indeed difficult to even start to begin thinking how this might be done). Instead, the focus in what follows will be on empirically pinning down the effect of the 9/11 and 7/7 attacks on subsequent patterns of hate crime incidence. Existing Evidence Linking Hate Crimes and Terrorist Attacks Quite a lot of descriptive evidence exists on whether terror attacks induce increases in hate crime. In the US it seems that the 9/11 terrorist attack caused an 7 increase in the number of hate crimes against Muslims, Arabs, and those perceived to be Middle Eastern.4 Consider the FBI annual statistics on hate crimes reported in Table 1 for the period 1997 to 2008. Prior to 2001, incidents of anti-Islamic crime were in the magnitude of 20-30 incidents per annum. This figure jumps from 28 incidents in 2000 to 481 in 2001, and then remains steady in the 100-150 range per annum thereafter. The total number of hate crimes committed (including racial, ethnic, sexual orientation and disability bias motivated crimes) remained approximately static during these ten years. Meanwhile, anti-Islamic crimes as a proportion of all hate crimes dramatically increased over this period. These FBI numbers convey the magnitude of the backlash against Muslims, but do not give any idea of the time-scale of the crimes within the year 2001 (as the figures are published annually). Since 9/11 occurred three quarters of the way through 2001, and the number of anti-Muslim hate crimes in the following year was significantly lower, it seems likely that the three months after the attacks were times of intense anti- Muslim violence. Other sources confirm this impression. Firstly, a report by the ADC (American- Arab Anti-Discrimination Committee) counts 700 violent attacks on US Muslims in the nine weeks following 9/11; they report that “the intensity of the backlash, especially in terms of hate crimes and discrimination, was at its peak in the first six months following the attacks, and particularly during the first nine weeks” (Ibish and Stewart, 2003, page 15). Secondly, Swahn, Mahendra, and Paulozzi (2003) conducted a survey of newspaper reports during the period 1st September 2001 - 11th October 2001. They found evidence of 100 incidents of hate crimes against Middle Easterners in the United 4 There is also some evidence that Sikhs were targeted, since the wearing of the turban was confused with the Arab headdress worn by Osama-bin-Laden and members of Al-Qaeda (Sheridan and Gillett, 2005). 8 States, of which just one occurred in the ten days between 1st September and 11th September (the “baseline”). Of the remaining 99, 77 occurred in the period ten days after 9/11. Incident types ranged from assault and intimidation to murder and attempted murder. Although this survey is not a rigorous scientific study (the baseline period is so short, and may be subject to seasonal variation), it does support the hypothesis of a relatively short and intense 'shock period'. What is more, this study gives direct evidence that the perpetrators of these hate crimes were motivated by the terrorist event: “the perpetrators in at least 30 of the incidents specifically mentioned the September 11 terrorist attacks, or accused the victims of being terrorists” (Swahn, Mahendra, and Paulozzi, 2003, page 188). Furthermore, there is some evidence to suggest that the effects of the 9/11 terrorist attacks was not limited to the United States. Surveys of Muslims in both the UK and Australia find a significant increase in experiences of hate crime post 9/11. In the UK, Sheridan and Gillett (2005) surveyed 398 respondents from various religious groups in Leicester and Stoke-on-Trent during the period October to December 2001. They estimate regressions to predict an aggregated 'change score' (showing change in experience of hate crimes since 9/11) and find that both Muslims and Hindus report increases post 9/11 (with a much larger effect for Muslims), while the other religious groups report small decreases. A similar, but smaller scale, study conducted in 2003 surveyed 186 Australian Muslims and Christians (Poynting and Noble, 2004) and found similar results, with Muslims being far more likely to report an increase in experiences of racism since September 11th than Christians. Thus there is survey evidence that the 9/11 terrorist attacks were immediately followed by a dramatic rise in the incidence of hate crimes against American Muslims, 9 with a peak lasting for around 2-3 months, and with the effects persisting for perhaps years afterwards. Other than the survey evidence already discussed, there exists little evidence of the experiences of British Muslims following 9/11. Even scarcer is evidence on the effect of 7/7, which we would presume likely to have caused similar effects to 9/11. Our empirical work will study the impact of both attacks. Evidence on the relationship between hate crimes and economic variables There is a serious shortage of convincing statistical evidence in this area. Gale et al. (2002) and Medoff (1999) do present tests of their theories which offer contrasting views on the role of economic factors on hate crimes. In the Gale et al model, relative well-being is important, meaning that an individual is particularly concerned with the hated group's situation relative to her own. Thus, if the hated group becomes comparatively better off in economic terms, they gain more utility from committing hate crimes against them. The authors test this theory using a panel of US states. They construct a fixed effects model using the hate crime rate as a dependent variable, and income per capita, the black/white income ratio, unemployment rate and relative sizes of minority populations as independent variables. Although few of these variables are found to be significant in the model, they do find that the black/white income variable is significant and supports the envy hypothesis – the higher are black wages relative to white wages, the higher are rates of hate crime against blacks. Moreover, the same model applied to all other crimes results in an oppositely signed coefficient on black/white income. Medoff (1999) tests his theory by estimating a cross-sectional state level model for the level of hate crimes as a function of the wage rate, average education level, unemployment rate and the proportion of the population aged 15-19 (since this age 10 group are likely to have the lowest value of time). The market wage is indeed found to have a significant negative relationship with wage, and unemployment a positive relationship – in states where citizens are more likely to be employed and earn more on average, the incidence of hate crime is lower. The findings of both these statistical analyses should, however, be considered with caution. It seems likely that in a cross-sectional empirical analysis such as this, results may be biased by omitted variables. Higher average wages may be correlated with omitted variables such as greater police presence and security (since wealthier states can afford to invest more in these) or with cultural differences; better paid and educated individuals may be more tolerant of other cultures. Since Medoff does not include the size of minority population in each state, this may cause bias. Ethnic minorities earn on average less than the majority population (Lindley, 2002) and so the negative association with wages may be due to the higher proportion of minorities in some states. More members of minority groups means both lower average wages and also more opportunities for individuals to perpetrate hate crimes. Gould and Klor (2012) consider an interesting dimension of possible longer run responses to hate crime increases induced by terror attacks. They study fertility and assimilation rates of immigrants from Muslim countries in the US post 9/11 and present some evidence that immigrants in US states that experienced sharper increases in hate crimes had higher probabilities of marrying in their own ethnic group, lower female employment and higher fertility. Thus, they argue that the terrorist attack of 9/11 induced a withdrawal back into communities by US muslims, thereby slowing down patterns of assimilation. 11 The other relevant piece of statistical work (that we are aware of) is Krueger and Pischke's (1997) examination of hate crimes in post-unification Germany (together with extensions of this work by Falk et al, 2011, and related work on extremism by Siedler, 2006). Their initial model finds a relationship between economic conditions and hate crimes, but once a dummy for East/West Germany has been added, and the model has been adjusted for censoring at zero (since a negative number of hate crimes is not possible) using a Tobit specification, the effect of unemployment disappears entirely. Krueger and Pischke conclude that it is the major social and attitudinal differences between East and West Germany that drive hate crimes, and that poor economic conditions are not a driver in and of themselves. 3. Data and Descriptive Analysis Data Data requirements to study the impact of terror attacks on hate crimes are very stringent. First of all, we need data on hate crimes measured in a consistent manner. We also need information on the ethnicity or religiosity of hate crime victims. Fortunately, for our purposes, data collected on hate crimes and on the victims of hate crimes by police forces in England is of very good quality owing to stringent definitional guidelines that police forces need to follow (see the Appendix on the nature of crime recording practices in England).5 Second, hate crime data at a high frequency (at least monthly) is required for us to carry out our empirical analysis of what happens to hate crimes before and after the 9/11 and 7/7 terror attacks. 5 The classification of hate crimes we use in data obtained from police forces in England seems to be collected on a much more systematic basis than the US data used in the research of Disha, Cavenish and King (2011), Deloughery, King and Asal (2012) and Gould and Klor (2012). 12 Data on monthly numbers of hate crimes broken down by ethnicity of victim is not publicly available. We therefore obtained such data by direct application to police forces through a freedom of information (FOI) request.6 We submitted FOI requests to four police force areas (PFAs) in England - the Metropolitan Police Services (MPS) in London, the West Midlands, Leicestershire and West Yorkshire. These were chosen for two main reasons. Firstly, because all four have a sizeable Muslim population, and thus hate crimes against Muslims are likely to occur relatively frequently, and secondly because the 7/7 attacks occurred in London, and so a comparison of London versus non- London areas was sought (the MPS covers all of central London, with the exception of City of London; the other three areas are independent of London (although, of course, the 7/7 bombers were from West Yorkshire). We obtained monthly data from all four police forces, with information being supplied to us on the major offence category and ethnicity for both victims and offenders of all crimes listed as racially motivated.7 For Leicestershire, London and the West Midlands we have data before and after both 9/11 and 7/7, and for West Yorkshire only for before and after 7/7. Significant crime recording changes occurred in April 2002 (see the Appendix for more detail) and this constrains us in our ability to look at before/after changes in 6 In the UK, a freedom of information request can be undertaken under the auspices of the 2001 Freedom of Information act. This gives individuals the right to ask any public body for all the information they have on a particular subject and, unless there is a good reason not to, the organization is required to supply the requested information. 7 Several caveats were attached to the data, describing recording changes, relevant events and sources of inaccuracy. Indeed, it is important to bear in mind that the data used in this study was not collected with our research purpose in mind, as expressed in the following words from West Midlands police force: “Every effort is made to ensure that the figures presented are accurate and complete. However, it is important to note that these data have been extracted from large disparate administrative data systems used by forces for police purposes. The detail collected to respond specifically to your request is subject to the inaccuracies inherent in any large scale recording system. As a consequence, care should be taken to ensure data collection processes and their inevitable limitations are taken into account when interpreting those data.” 13 hate crime associated with the two terror attacks. In fact, it means the feasible time series we can study differ around the window of the two attacks. We can do a much better job on having consistent data before and after 7/7 and so our main focus is placed on studying what happened to hate crimes in response to this terror attack. We thus study the 7/7 attacks first and then look at 9/11 effects using a shorter time series that stops when the recording changes occurred in April 2002. The actual periods studied are as follows: 7/7 attack – January 2003 to December 2007; 9/11 attack – April 2000 to March 2002. Hate Crimes by Ethnicity There is a distinction between racial and religious discrimination, although often the two co-occur. While it is clear that the 9/11 and 7/7 terrorist attacks triggered animosity towards Western Muslims, research from the US (discussed previously) has found that it is not just Muslims who were targeted – hate crimes were also carried out on Middle Easterners and Arabs who were not practising Muslims, and Sikhs, who were mistaken for Muslims. The vast majority of Britain's Muslim population are South Asian, most of whom originate from Pakistan, Bangladesh and India. The Pakistani and Bangladeshi populations are almost entirely Muslim, while the Indian population sub-divides into Hindus, Muslims and Sikhs. Thus the UK Muslim population is almost entirely contained within the ethnic category 'South Asian'. People in this category may be the victim of either racial or religious discrimination. In some cases, religious discrimination may be misplaced – individuals may be discriminated against because they are mistaken for being Muslims, or because of some kind of statistical profiling 14 (i.e. discriminators target South Asians because they are the ethnic group most likely to contain Muslims). Since religious data was unavailable, we use ethnicity to define our main groups of interest. Ethnicity categories used in crime statistics differ from one police force to the next, and so some aggregation was required in order to standardise the figures from the different sources. The following six broad categories were created: Asian/Arab, White, Black, Oriental, Unknown, Other. The latter three contain small numbers and so are generally dismissed from analysis. We thus consider the impact of the terror attacks on Asian/Arab hate crimes and use hate crimes against Whites and Blacks as control groups when we formulate our statistical models. Descriptive Analysis The vast majority of hate crimes involve violence against the person. Table 2 shows the major offence categories (excluding sexual offences) of all hate crimes against Asians/Arabs between January 1998 and March 2010 from our FOI request to the Metropolitan Police Service. Over three quarters (77 percent) were classified as violence against the person, and a further 18 percent involved criminal damage. Figure 1 plots the monthly time series of hate crimes by ethnic group and police force area for the time window for which we study the 7/7 attacks (January 2003 to December 2007). There are several interesting features of the overall patterns. First, whilst the monthly time series do jump around, all four graphs show a discernible spike up in the Arab/Asian hate crime series in the month of July 2005, suggesting an immediate impact. Second, eyeballing the graphs is suggestive of the notion that the time series patterns of hate crimes before the 7/7 bombings for all three ethnic groups look similar (this is considered formally in more detail below). 15 There are also two police force area specific observations that are relevant: i) In the West Midlands there is a large peak caused by the Birmingham race riots in October 2005, which were sparked by the alleged rape of a Black girl by a group of South Asian men. This event seems to have been completely unrelated to the terrorist attacks that occurred three months previously. ii) The pre-recording change data for West Yorkshire was not good enough to study the 9/11 attacks for this police force area. Also, they introduced a True Vision third party recording scheme was launched in June 2005, just one month before 7/7. We deal with these two data issues in our empirical models below by including specific variables to control for any data jumps from these. An analogous set of charts for a shorter time window around the 9/11 attacks (April 2000 to March 2002) is given in Figure 2. The chart this time covers only three police force areas excluding West Yorkshire. Whilst the length of the post-attack time period is constrained by the recording changes of April 2002, the Figure does seem to show a blip up in hate crimes against Asians/Arab 9/11 and higher relative levels (despite subsequent falls) compared to the White and Black hate crimes. We scrutinise these patterns in more detail in the statistical models in the next Section of the paper. 4. Modelling Approach and Statistical Results Basic Approach We begin the statistical analysis by developing and empirical model that permits us to study the question of how the 7/7 and 9/11 terror attacks impacted upon hate crime. We ask what happened to hate crime against Asians and Arabs before and after the terror attacks relative to hate crime against two other ethnic groups (Blacks and Whites). 16 Because crime is seasonally highly persistent8, and we use monthly data across years, we express our model in twelve month differences (thereby differencing out area and month fixed effects from a levels model). We operationalise our estimator in terms of the following equation determining twelve month changes in hate crimes against Asians and Arabs HA in area j in month m in year t: A W W B B 12 jmt 12 jmt 12 jmt mt t jmtH = δ H +δ H + θA + τ + ε∆ ∆ ∆ (1) where HW denotes hate crimes against Whites, HB denotes hate crimes against Blacks, τt is a set of year dummies, εjmt is an error term and the main variable of interest Amt is a dummy variable equal to one in months where the terror attack occurred (or for a window comprising several post-attack months - see below) and zero otherwise. ∆12 is a twelve month differencing operator. The inclusion of the two other hate crime variables (for Whites W and Blacks B) enables us to ascertain the impact of terror attacks on the Asian and Arab group relative to these groups. Thus, in (1) the attack coefficient θ estimates whether Asian/Arab hate crimes differentially increased when the terror attacks occurred and, when Amt is defined to cover a longer post-attack duration, how they evolved subsequently in the wake of terror attacks. Pre-Attack Trends A prerequisite for this estimator to yield unbiased estimates is that pre-attack trends of hate crimes against the treatment group (Asians/Arabs) are no different to trends in hate crimes against the comparison groups (Whites and Blacks). A glance back to Figures 1 and 2 makes it graphically clear how this operates in practice, as the 8 See Hird and Ruparel (2007) on the seasonality of crime or Draca, Machin and Witt (2011) who difference weekly crime data across years, among others. 17 Asian/Arab, White and Black hate crime trends do seem to show strong similarities in pre-attack periods. This formally tested for the 7/7 attacks in Table 3. The upper Panel of the Table considers pre-attack trends in twelve month differenced hate crimes for all four areas pooled for hate crimes against Asians/Arabs (in specification (1)), Whites (in (2)) and Blacks (in (3)). In specification (4) we condition the Asian/Arab hate crimes on the control group crimes. The lower panel reports the same specifications where we also allow for area-specific trends. The estimated coefficients on the trend variables in both Panels show there to be no differential pre-attack trends between Asian/Arab hate crimes and those against Whites and Blacks. Thus, the common trends assumption required for our estimator to be valid is upheld. Estimates of the 7/7 Impact Table 4 shows estimates of equation (1) for the 7/7 study time period. There are four Panels in the Table, where each gives a 7/7 impact over different durations. Panel A shows the immediate impact via a dummy variable defined for the 7/7 month only. Panels B to D further refine the dummy variable definition to cover a wider post-attack window (Panels B, C and D respectively refine the dummy variables to cover three, six and twelve months post-attack). Results from three specifications are included in each Panel. The first includes only the 7/7 dummy in the twelve month differenced model (which also contains year dummies). The second includes the twelve month differenced White and Black hate crime variables. The third additionally conditions on the total crime rate, again differenced across months in adjacent years. All the specifications are population 18 weighted and report Newey-West standard errors (with lag t-12 owing to the twelve month differencing). Consider first the immediate impact results in Panel A of Table 4. Specification (1) produces a 0.33 coefficient on the 7/7 dummy, showing a significant spike up of around one third in hate crimes against Asians/Arabs in the attack month.9 Specification (2) shows the estimate that conditions out the White and Black hate crimes, and the estimated coefficient remains strongly significant, but falls to 0.23. The estimated coefficients on the White and Black hate crime variables are positive and significant, showing the need to control for them and the fact that they covary positively over time with the Asian/Arab hate crime variable. The third specification adds in the total crime rate, which itself attracts a positive coefficient (which is significant at the 10 percent level, with a p-value of 0.09), but barely changes the estimated 7/7 impact. In this specification hate crimes against Asians/Arabs rose by 22 percent in the attack month. Panel B considers impact in the three months following the terror attacks. The estimated impact comes down, but remains strong and significant at 16 percent in the full model. The window is further widened in Panels C and D where the effects again fall but remain significant. The magnitudes are 7.2 percent after six months and 8.5 percent after twelve. The results of Table 4 show a strong impact of 7/7 on hate crimes against Asians and Arabs. The immediate impact is largest, but the effect persists even twelve months after the attack occurred. Four specific estimates were chosen to be reported. We can, however, estimate an impact for every month sequentially to study the duration of 9 We report t-12 Newey-West standard errors since we study monthly time series data. However, we have seasonally differences the data relative to the same month in adjacent years, so may need to consider a longer lag length. Reassuringly, use of lag t-24 standard errors did not alter results by much. For example, the estimated standard error for the column (1) Panel A 7/7 coefficient was 0.063 for the t-24 specification (as compared to 0.071 for t-12). 19 impact in more detail. Estimated coefficients (and associated 95 percent confidence intervals) from carrying out this sequential modelling exercise are given in Figure 3, which reports estimates for up to 29 months after the 7/7 attacks. The estimates reproduce the large immediate impact at 22 percent in July 2005, which falls to 5 percent if the window is defined as four months after the attack. After that it stabilises in the range of 5 to 10 percent higher, all the way to the end of the sample (the 29 month estimate is up to December 2007 and therefore compares the whole post-attack period to the whole pre-attack period). The estimates are mostly significantly different from zero, though sometimes just drop beneath the 5 percent statistical significance level. In fact for the 25 estimates from 5 months out to 29 months out, 11 are significant at the 5 percent level, and 14 just drop. If a more conservative 10 percent significance level were used, 17 out of 25 are significant. The results in Table 4 and Figure 3 are very supportive of the idea that 7/7 caused a strong immediate increase in hate crimes against Asians and Arabs, and that whilst the scale of the increase tempered off through time, they remained around 5 to 10 percent higher than the pre-attack levels. Thus there seems to have been a permanent impact of increasing hate crimes against Asians and Arabs that occurred in the wake of the 7/7 bombings. Consideration of Economic Factors Some of the small body of empirical work on hate crimes (e.g. Gale et al., 2002; Medoff, 1999; Krueger and Pischke, 1997) have considered a role for economic outcomes to influence hate crime. We look at this in Table 5 by adding economic variables (taken from the British Crime Survey at monthly frequency for the four police force areas) to the most detailed Table 4 specification. The three economic variables 20 considered are log(income), the proportion of the population with no educational qualifications and the proportion in work. Each of these attracts an insignificant coefficient in all four specifications reported. Thus, differential economic conditions do not seem to be key predictors of hate crimes against Asians and Arabs. Moreover, the 7/7 impact is almost entirely unaffected by the inclusion of these economic factors. The increase in hate crimes against Asians and Arabs that occurred, and persisted, after 7/7 is not due to economic conditions worsening after the attacks occurred. Separate Estimates by Police Force Area In Table 6 we report separate estimates of the 7/7 impact, at the different chosen post-attack durations, by police force area. There are at least two reasons for doing this. First, as highlighted above, there are certain police force area specific data issues of relevance. Second, we wish to explore possible heterogeneities in the magnitude and duration of impact across areas. The Table confirms there to be some variation. In terms of immediate impact, there is a significant impact in London (of 22 percent impact on Log(Asian/Arab hate crimes)), Leicestershire (of 30 percent) and West Yorkshire (of 42 percent). The West Midlands impact is not significant. It does, however, become significant if we consider a three month period after the attack, as is the case in all four areas. The rate of decay of these effects differs by area with, interestingly, there being no impact remaining in London twelve months after the terror attacks, but the effects still persisting strongly in the other three police force areas. Estimates of the 9/11 Impact We have also estimated versions of equation (1) for the impact of 9/11. However, we should say that this analysis is much more limited than for the study of the 21 7/7 impact. There are several dimensions to this. First, as noted above, we only have usable data for three police force areas. Second, we are not able to define a symmetric time series window around the attack as we did with the 7/7 analysis. This is because we have to stop due to the recording practice change that occurred in April 2002. One consequence of this is that we cannot look for longer run impacts; we can only look as far as six months after the terror attack. Third, we do not have data on the overall crime rate in the police force areas as we only have that on a consistent basis at monthly frequency after the recording practice change. Fourth, because we only have two years’ data, we can difference across months in the years but are not able to calculate Newey- West standard errors with full lag length t-12 and have to instead use a t-1 structure instead. The results are reported in Table 7. The Table is structured in a comparable way to the 7/7 results, though we can only look at shorter duration effects. The column (1) results show a strong immediate impact effect from 9/11. Hate crimes against Asians and Arabs rose by 28 percent in September 2001. This effect dampens down by three months after the attacks to 22 percent, and falls further to 11 percent after six months. Figure 4 shows estimates for every month post-attack duration as with the earlier 7/7 analysis. Whilst it is only possible to study a shorter duration, a rather similar pattern emerges, with a strong positive initial impact which dies down but appears to settle at a higher level than the pre-attack period. Out of seven estimated coefficients, five are significant at the 5 percent level, and all are significant at a 10 percent level of significance. 22 5. Conclusions Despite the importance of the subject, credible statistical evidence on the impact of terror attacks on hate crime is hard to come by. In this paper we look at the impact of the 7/7 and 9/11 terrorist attacks on hate crimes against Asians and Arabs in four police force areas of the UK. We estimate a strong immediate impact on Asian/Arab hate crimes from both terror attacks, find that the effects of both persist and that hate crimes move to a permanently higher post-attack level. The highly similar pattern of results from the separate study of the respective impacts of 7/7 and 9/11 on hate crime in four areas with sizable Asian/Arab populations is highly suggestive that a causal impact of terror attacks on hate crime is identified from the empirical approach implemented in the paper. These findings add to the literature on the economic and social effects of terror attacks. They show, in line with some of the theoretical discussion in the early part of the paper, that in terms of individuals the cost of terror attacks is not just limited to the victims of the attacks. The fact that hate crimes perpetrated against Asians and Arabs significantly rose in the wake of 9/11 and 7/7 points to an additional social cost of terrorist activity. If attitudes towards groups like British Muslims are altered by attacks and by media coverage of attacks then these findings certainly fit with the proposition of 'attitudinal shocks', where a driver of hate crimes is the level of hatred or bigotry about a particular group in society, which may well be influenced by media coverage of attacks. In this setting, such shifts in underlying bigotry from attitudinal change following events like terrorist attacks seem more important as determinants of hate crime incidence than do socio-economic factors. Thus, the determinants of hate crimes may well be different from the kind of incentive effects or deterrence effects that 23 emerge as crime determinants in the standard economics of crime model. Of course, more work on the causes of hate crime and on the behavioural motives that individuals have to engage in crime against different ethnic or religious groups forms an important future research agenda. 24 References Abadie, A. and J. Gardeazabal (2003) The Economic Costs of Conflict: A Case Study of the Basque Country, American Economic Review, 93, 113-132. Baddeley, M. (2010) Herding, Social Influence and Economic Decision-Making: Socio- Psychological and Neuroscientific Analyses, Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 365, 281-290. Becker, G. (1957) The Economics of Discrimination (2nd ed., 1.), Chicago: Univ. Chicago Press. Becker, G. (1968) Crime and Punishment: An Economic Approach, The Journal of Political Economy, 76, 169-217. Becker, G., and Y. Rubinstein (2009) Fear and the Response to Terrorism: An Economic Analysis, mimeo. Berman, G. (2008) Changes in Crime Recording Practices, Technical report, Library of the House of Commons, Standard Note SN/SG/2607. Bloom, N. (2009) The Impact of Uncertainty Shocks, Econometrica, 77, 623-685. Bozzoli, C. and C. Mueller (2009). Perceptions and Attitudes to a Terrorist Shock: Evidence from the UK, Economics of Security Working Paper 13. Chen, A. And T. Siems (2004) The Effects of Terrorism on Global Capital Markets, European Journal of Political Economy, 20, 349-366. Craig, K. (2002) Examining Hate-Motivated Aggression: A Review of the Social Psychological Literature on Hate Crimes as a Distinct Form of Aggression, Aggression and Violent Behaviour, 7, 85-101. Deloughery, K., R. King and V. Asal (2012) Close Cousins or Distant Relatives? The Relationship Between Terrorism and Hate Crime, Crime and Delinquency, forthcoming. Disha, I., J. Cavendish and R. King (2011) Historical Events and Spaces of Hate: Hate Crimes Against Arabs and Muslims in Post 9/11 America, Social Problems, 58, 21-46. Draca, M., S. Machin and R. Witt (2011) Panic on the Streets of London: Crime, Police and the July 2005 Terror Attacks, American Economic Review, 101, 2157-81. Eskenazi, B., A. Marks, R. Catalano, T. Bruckner and P. Toniolo (2007) Low Birthweight in New York City and Upstate New York Following the Events of September 11th, Human Reproduction, 22, 3013-3020. 25 Falk, A., A. Kuhn and J. Zweimuller (2011) Unemployment and Right Wing Extremist Crime, Scandinavian Journal of Economics, 113, 260-85. Frey, B., S. Luechinger and A. Stutzer (2004) Valuing Public Goods: The Life Satisfaction Approach, CESifo Working Paper Series. Gale, L., W. Heath and R. Ressler (2002) An Economic Analysis of Hate Crime, Eastern Economic Journal, 28, 203-216. Gould, E. And E. Klor (2012) The Long-Run Effect of 9/11: Terrorism, Backlash, and the Assimilation of Muslim Immigrants in the West, CEPR Discussion Paper 8797. Gordon, C. and A. Arian (2001) Threat and Decision Making, Journal of Conflict Resolution, 45, 196-215. Hird, C. And C. Ruparel (2007) Seasonality in Recorded Crime: Preliminary Findings, Home Office Online Report 02/07. Ibish, H. and A. Stewart (2003) Report on Hate Crimes and Discrimination Against Arab Americans, October, retrieved from http://www.adc.org/PDF/hcr02.pdf. Krueger, A. and S. Pischke (1997) A Statistical Analysis of Crime Against Foreigners in Unified Germany, The Journal of Human Resources, 32, 182-209. Lauderdale, D. (2006) Birth Outcomes for Arabic-Named Women in California Before and After September 11, Demography, 43, 185-201. Lindley, J. (2002) Race or Religion? The Impact of Religion on the Employment and Earnings of Britain's Ethnic Communities, Journal of Ethnic and Migration Studies, 28, 427-42. Medoff, M. (1999) Allocation of Time and Hateful Behavior : A Theoretical Analysis of Hate and Hate Crimes, American Journal of Economics and Sociology, 58, 959-973. Metcalfe, R., N. Powdthavee and P. Dolan (2011) Destruction and Distress: Using a Quasi-Experiment to Show the Effects of the September 11 Attacks on Mental Well-Being in the United Kingdom, Economic Journal, 121, F81-103. Poynting, S. And G. Noble (2004) Living with Racism: The Experience and Reporting by Arab and Muslim Australians of Discrimination, Abuse and Violence Since 11 September 2001, Sydney Human Rights and Equal Opportunities Commission. Sheridan, L. And R. Gillett (2005) Major World Events and Discrimination, Asian Journal of Psychology, 8, 191-197. 26 Siedler, T. (2006) Family and Politcs: Does Parental Unemployment Cause Right Wing Extremism, IZA Discussion Paper 2411. Simmons, J., C. Legg and R. Hosking (2003) National Crime Recording Standards (NCRS): An Analysis of the Impact on Recorded Crime; Part Two: Impact on Individual Police Forces, Home Office Online report 32/03. Smits, L., L. Krabbendam, R. de Bie, G. Essed, and J. van Os (2006) Lower Birth Weight of Dutch Neonates who were in Utero at the Time of the 9/11 Attacks, Journal of Psychosomatic Research, 61, 715-717. Sunstein, C. (2003) Terrorism and Probability Neglect, Journal of Risk and Uncertainty, 26, 121-136. Swahn, M., R. Mahendra and L. Paulozzi (2003) Violent Attacks on Middle Easterners in the United States During the Month Following the September 11, 2001 Terrorist Attacks, Injury Prevention, 9, 187-189. Viscusi, K. And R. Zeckhauser (2003) Sacrificing Civil Liberties to Reduce Terrorism Risks, Journal of Risk and Uncertainty, 26, 99-120. 27 Figure 1: Trends in Hate Crimes by Ethnicity of Victim, Four Police Force Areas, January 2003 to December 2007 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 H a te C ri m e s Jan-03 7/7 Dec-07 Month Asian/Arab White Black London 0 1 0 0 2 0 0 3 0 0 H a te C ri m e s Jan-03 7/7 Dec-07 Month Asian/Arab White Black West Midlands 0 1 0 0 H a te C ri m e s Jan-03 7/7 Dec-07 Month Asian/Arab White Black Leicestershire 0 1 0 0 2 0 0 H a te C ri m e s Jan-03 7/7 Dec-07 Month Asian/Arab White Black West Yorkshire 28 Figure 2: Trends in Hate Crimes by Ethnicity of Victim, Three Police Force Areas, April 2000 to March 2002 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 H a te C ri m e s Apr-00 9/11 Mar-02 Month Asian/Arab White Black London 0 1 0 0 2 0 0 3 0 0 H a te C ri m e s Apr-00 9/11 Mar-02 Month Asian/Arab White Black West Midlands 0 1 0 0 H a te C ri m e s Apr-00 9/11 Mar-02 Month Asian/Arab White Black Leicestershire 29 Figure 3: 7/7 Impact - Time Varying Coefficients With Effects Up to December 2007 -. 1 -. 0 5 0 .0 5 .1 .1 5 .2 .2 5 .3 .3 5 E st im a te d C o e ff ic ie n t a n d 9 5 % C I 0 5 10 15 20 25 30 Months Since 7/7 Notes: Estimated coefficient and 95% confidence interval from the column (3) specification in Table 4. 30 Figure 4: 9/11 Impact - Time Varying Coefficients With Effects Up to March 2002 0 .1 .2 .3 .4 .5 E st im a te d C o e ff ic ie n t a n d C I 0 1 2 3 4 5 6 Months Since 9/11 Notes: Estimated coefficient and 95% confidence interval from the Panel A specification in Table 7. 31 Table 1: US Hate Crime Statistics Year Total Hate Crimes Reported Offenders' Reported Motivations Religious Bias Anti-Islamic 1997 8049 1385 28 1998 7755 1390 21 1999 7876 1411 32 2000 8063 1472 28 2001 9730 1828 481 2002 7462 1426 155 2003 7489 1343 149 2004 7649 1374 156 2005 7163 1227 128 2006 7722 1462 156 2007 7624 1400 115 2008 7783 1519 105 Notes: From the Anti-Defamation League's Washington Office based upon FBI information (www.adl.org). 32 Table 2: Major Offence Categories For Hate Crimes Committed Against Asians/Arabs in London (January 1998 to March 2010) Offence Category Total Hate Crimes Reported Percentage Violence against the person 45078 76.9 Criminal damage 10605 18.1 Robbery 908 1.5 Theft and handling 872 1.5 Other notifiable offences 821 1.4 Burglary 240 0.4 Fraud or forgery 66 0.1 Sexual offences censored censored Total 58590 100 Notes: From Metropolitan Police Service records supplied in freedom of information request. 33 Table 3: Pre-7/7 Trends in Hate Crimes Against Asians/Arabs, Whites and Blacks (January 2003 to June 2005) Log(Asian/Arab Hate Crimes) Log(White Hate Crimes) Log(Black Hate Crimes) Log(Asian/Arab Hate Crimes) (1) (2) (3) (4) A. Four Areas Pooled Trend -0.001 (0.008) -0.003 (0.007) -0.008 (0.010) 0.002 (0.005) Log(White Hate Crimes) 0.373 (0.113) Log(Black Hate Crimes) 0.103 (0.071) B. Area Specific Trends Trend X London -0.002 (0.003) -0.003 (0.006) -0.010 (0.006) 0.001 (0.004) Trend X West Midlands -0.001 (0.003) -0.004 (0.006) -0.008 (0.005) 0.001 (0.004) Trend X Leicestershire 0.001 (0.003) -0.001 (0.006) -0.006 (0.006) -0.000 (0.004) Trend X West Yorkshire 0.003 (0.003) -0.001 (0.006) -0.003 (0.006) 0.004 (0.003) Log(White Hate Crimes) 0.373 (0.113) Log(Black Hate Crimes) 0.103 (0.071) Notes: All models estimated on monthly data across four police force areas from January 2003 to June 2005; Population weighted; Seasonally differenced across the same months in adjacent years; Newey West standard errors (lag t-12) in parentheses. 34 Table 4: Hate Crimes Against Asians/Arabs and the 7/7 Terror Attacks (January 2003 to December 2007) Log(Asian/Arab Hate Crimes) (1) (2) (3) A. Month of Terror Attack 7/7 .330 (.071) .231 (.053) .221 (.055) Log(White Hate Crimes) .397 (.058) .379 (.054) Log(Black Hate Crimes) .287 (.069) .288 (.064) Log(Total Crime) .415 (.240) B. 3 Months From Terror Attack 7/7 .180 (.050) .171 (.033) .158 (.036) Log(White Hate Crimes) .411 (.058) .394 (.054) Log(Black Hate Crimes) .288 (.067) .289 (.064) Log(Total Crime) .370 (.242) C. 6 Months From Terror Attack 7/7 .030 (.080) .090 (.036) .072 (.038) Log(White Hate Crimes) .419 (.059) .398 (.055) Log(Black Hate Crimes) .296 (.068) .296 (.065) Log(Total Crime) .386 (.248) D. 12 Months From Terror Attack 7/7 .070 (.074) .095 (.041) .085 (.042) Log(White Hate Crimes) .430 (.060) .411 (.055) Log(Black Hate Crimes) .283 (.067) .285 (.064) Log(Total Crime) .366 (.246) Notes: All models estimated on monthly data across four police force areas from January 2003 to December 2007; Population weighted; Seasonally differenced across the same months in adjacent years; Newey West standard errors (lag t-12) in parentheses; Year dummies, a dummy for the Birmingham race riot in October 2005 and for the introduction of the True Vision recording scheme in West Yorkshire from June 2005 onwards are included in all specifications. 35 Table 5: Consideration of Economic Factors Log(Asian/Arab Hate Crimes) (1) (2) (3) (4) Month of Terror Attack 3 Months From Terror Attack 6 Months From Terror Attack 12 Months From Terror Attack 7/7 0.225 (0.056) 0.165 (0.037) 0.082 (0.039) 0.098 (0.044) Log(White Hate Crimes) 0.375 (0.056) 0.391 (0.057) 0.371 (0.052) 0.410 (0.059) Log(Black Hate Crimes) 0.278 (0.066) 0.278 (0.067) 0.286 (0.067) 0.270 (0.066) Log(Total Crime) 0.437 (0.264) 0.395 (0.265) 0.405 (0.271) 0.375 (0.270) Log(Income) 0.073 (0.121) 0.073 (0.119) 0.085 (0.119) 0.101 (0.114) Proportion No Qualifications -0.119 (0.163) -0.123 (0.167) -0.103 (0.178) -0.170 (0.188) Proportion in Employment -0.255 (0.221) -0.293 (0.219) -0.288 (0.212) -0.310 (0.214) Notes: As for Table 4. Monthly data on Income, No Qualifications and Employment by police force area and month from the British Crime Survey. See the Appendix for more detail. 36 Table 6: Separate 7/7 Impact Estimates by Police Force Area Log(Asian/Arab Hate Crimes) (1) (2) (3) (4) Month of Terror Attack 3 Months From Terror Attack 6 Months From Terror Attack 12 Months From Terror Attack 7/7 X London 0.217 (0.019) 0.151 (0.025) 0.048 (0.042) 0.021 (0.031) 7/7 X West Midlands 0.043 (0.051) 0.072 (0.036) 0.156 (0.052) 0.170 (0.041) 7/7 X Leicestershire 0.299 (0.022) 0.268 (0.022) 0.101 (0.044) 0.177 (0.046) 7/7 X West Yorkshire 0.416 (0.046) 0.239 (0.057) 0.051 (0.083) 0.187 (0.080) Log(White Hate Crimes) 0.375 (0.054) 0.387 (0.055) 0.394 (0.057) 0.378 (0.059) Log(Black Hate Crimes) 0.291 (0.067) 0.290 (0.068) 0.288 (0.067) 0.250 (0.065) Log(Total Crime) 0.417 (0.243) 0.378 (0.248) 0.383 (0.256) 0.353 (0.257) Notes: As for Table 4. 37 Table 7: Hate Crimes Against Asians/Arabs and the 9/11 Terror Attacks (April 2000 to March 2002) Log(Asian/Arab Hate Crimes) (1) (2) (3) Month of Terror Attack 3 Months From Terror Attack 6 Months From Terror Attack 9/11 0.280 (0.082) 0.223 (0.066) 0.109 (0.070) Log(White Hate Crimes) 0.399 (0.107) 0.357 (0.092) 0.386 (0.102) Log(Black Hate Crimes) -0.060 (0.099) -0.108 (0.099) -0.041 (0.015) Notes: All models estimated on monthly data across three police force areas from April 2000 to March 2002; Population weighted; Seasonally differenced across the same months in adjacent years; Newey West standard errors (lag t-1) in parentheses; Year dummies are included in all specifications. 38 Appendix Changes in Recording Practices The crime data we obtained from four police force areas in a Freedom of Information request sent in the Summer of 2010 are potentially subject to various recording changes that occurred both locally and nationally. There were two important national changes in crime recording during the time period of study (Berman, 2008). These affect the start and end dates of our study periods and mean that we are forced to carry out our 9/11 and 7/7 analyses for different time periods. The two changes are: i) The first was a change to Home Office counting rules which occurred in 1998/1999. New crime types were recorded in crime statistics for the first time; minor criminal damage was recorded where before it had not been, and there was a shift towards counting one crime per victim, rather than per offender. The result was a significant increase in the number of crimes recorded. The change affected different crime types and areas differently. Drug offences and violent crime saw the greatest increases as a result of the new counting rules. All of the police forces, with the exception of the MPS, warned that data prior to February 2000 either lacked accuracy or was different due to changes in Home Office counting rules. For this reason, we begin any potential analysis in the financial year beginning April 2000. ii) The second major change was this introduction of the National Crime Recording Standard (NCRS) in April 2002. The purpose of this was to standardise crime recording practices across police forces to allow between-force comparisons and to generate a better estimate of the national crime level. The purpose of the NCRS was also to move towards victim-focused crime recording, with 'victimless crimes' not being recorded under the new guidelines. The total impact of the NCRS is estimated to be approximately a 10% increase in crime in 2002/3 over the pre-NCRS level. However, different offence categories were affected differently – the greatest increase was in violence against the person, which is estimated to have increased by 23% nationally after introduction of the NCRS (Berman, 2008). The majority of hate crime falls into this category (as shown in Tables 2 of the paper - 77% of hate crimes are 'violence against the person', compared with 19% of all crime), and so the effects of the NCRS on hate crime are likely to be significant. Analysis of the effects of the NCRS on individual police-forces can be found in Simmons, Legg and Hosking (2003). Leicestershire adopted the NCRS in April 2002, but also adopted a Centralised Crime Recording Bureau in August 2002. An increase in the number of crimes recorded, especially violent crimes, has been noted since the introduction of the NCRS, but the size of the effect cannot be untangled from the effect of the recording change which occurred several months later. The MPS also adopted the NCRS in April 2002. It estimates the effect on all crime for the financial year 2002/3 to be approximately 12%, and 20% for violence against the person. The West Midlands adopted the NCRS in January 1999, along with the revised Home Office counting rules and so there is no change in trend for 2002/3. Owing to the introduction of the NCRS we stop our analysis of 9/11 in March 2002. Our 7/7 analysis runs from January 2003 to December 2007 and so is unaffected. 39 In terms of local changes in recording practices, West Yorkshire adopted the NCRS in February 2002 and experienced the largest effect, with an estimated 47% NCRS impact on violence against the person for the year 2002/3. In addition, to the national recording changes, West Yorkshire also implemented some regional recording and reporting changes during the period studied. These were documented in the FOI response as follows: 1) 1999: The MacPherson Report introduced a new definition of what constituted a racist crime/incident. 2) 2000: policing divisions in West Yorkshire appointed hate crime co-ordinators 3) December 2003: Introduction of a Vulnerable & Intimidated Victim Database (VIVID) to record and monitor incidents of domestic violence, hate and child abuse 4) June 2005: West Yorkshire Police signed up to the True Vision third party recording scheme for hate crimes. As a result we excluded West Yorkshire from the 9/11 analysis study period. British Crime Survey Data For the 7/7 analysis study period we matched in data by month and police force area from the British Crime Surveys (of 2002/3, 2003/4, 2004/5, 2006/7 an 2007/8) on income, the proportion with no educational qualifications and the proportion in work. British Crime Survey data is used here (rather than, for example, Labour Force Survey data) as sample sizes are large at around 45,000 individuals per year and there is an area variable (not available in other surveys) which perfectly matches police force areas.